Chronological Age Determination for Forensic Applications using Random Forest Regression and DNA Methylation Analysis
Friday, 19.1.18, 12:00-13:00, Raum 404, Eckerstr. 1
Over the last few years it became clear that additional information is hidden wi-\nthin epigenetic modifications of the DNA, and that especially DNA methylation\n(DNAm) could provide useful evidence to the criminal justice system. Within this\nproject, specific changes in DNAm levels upon age progression at selected loci were\nused to develop an objective scientific tool to determine the chronological age of\nan (unknown) individual. This information can be used to narrow down the list\nof suspects during criminal investigations or to determine the age of a person in\nother legal contexts such as human trafficking. A model for age prediction based\non whole blood samples, 13 selected age-dependent DNAm markers, and a ran-\ndom forest regression (RFR) approach was developed. The analysis of the DNAm\nwas performed using amplicon based massive parallel sequencing (MPS) and the\nRFR model created with the R package RandomForest. The performance of the\nmodel was evaluated using cross-validation for the training set and by indepen-\ndent analysis of an additional test set. Within the seminar, a short introduction\ninto the field of forensic (epi-)genetics, the marker selection and development of\nthe DNA methylation tool based on RFR and MPS as well as the results of the\nage-determination tool will be presented. Furthermore, the potential and (current)\nlimitations of the experimental and machine learning approach in respect to the\nimplementation into forensic investigations will be discussed. The here presented\nproject of the University of Amsterdam in cooperation with the Netherlands Fo-\nrensic Institute was funded by the NCTV grant of the Dutch Ministry of Security\nand Justice.
Higher Order Elicitability
Friday, 26.1.18, 12:00-13:00, Raum 404, Eckerstr. 1
Elicitability of a statistical functional means that it can be obtained as the minimizer of an expected loss function. Such a loss function leads to a natural way of forecast comparison or model selection, and allows for M-estimation and generalized regression.\n\nPrime examples of elicitable functionals are the mean or quantiles of a random variable. Independently, Weber (2006, Mathematical Finance) and Gneiting (2011, JASA) have shown that expected shortfall (ES), an important risk measure in banking and finance, is not elicitable. However, it turns out that ES is jointly elicitable with a certain quantile, that is, it is elicitable of second order.\n\nIn this talk, we present our results on higher order elicitability of ES and some other functionals, and we provide characterizations of the associated classes of consistent scoring functions. We illustrate the usefulness of scoring functions for forecast comparison.